Evaluating the Performance of Deep Learning Models in Predicting Brain Ischemic Stroke Risk using EEG Signals.

Stroke is the third most common cause of fatalities worldwide. It’s a hazardous condition that may cut off blood flow to the brain and result in death, serious illness, or disability. In order to begin treatment early and reduce the death rate, an exact stroke prediction is essential. Electrical activity inside the brain is reflected in electroencephalography (EEG) data, which provide a non-invasive method of monitoring brain performing to detect abnormalities relating to stroke risk. The intention is to evaluate the way deep learning models predict the risk brain stroke using EEG signals. For training and validate the models, a large set of electroencephalogram (EEG) data from patients with different stroke risk levels was used Using neural network architectures, which include recurrent and convolutional neural networks (RNNs), intends to identify EEG data’s temporal and spatial features that are predictive of the emergence of a stroke.At the convolutional layer, preprocessed data is fed into the three-layer CNN-model. From the input data, the layer of convolution extracts features that are useful. after feature extraction, fully interconnected (FC) layers are used to process the features by changing their weights. RNNs are made to handle sequential data by keeping memory states, particularly LSTM (Long Short-Term Memory) networks. RNN excels at recognizing sequential data’s In order to take benefit of both architectures, a hybrid approach which utilizes CNN for extraction of features and RNN for modeling sequences can be implemented. In order to simulate temporal dependencies, layers of CNN may first extract spatial information, which are then given to RNN layers. The hybrid algorithm achieved $\mathbf{9 0. 0 2 \%}$ accuracy.